How a governed request works

Your AI. Your rules.

Define the work.
Govern the context.
Inspect the answer.

The chosen reference standard

Lectotype puts your data, rules and permissions in charge before the model ever sees a prompt.

Who decides what the model sees?

Typical agent same question

DataRulesHistory

LLM

The model picks its own context and next steps.

With Lectotype same question

DataRulesHistory

Identity and permitted scope

Eligible evidence and rules

Prompt built by the rules

LLM

The model gets one defined language task.

Same question. The difference is where the decisions live: in rules you control, outside the model.

Every request shows its credentials.

Who is asking?Identity and privileges

What is the job?Scope and requirements

Is the evidence sound?Quality and provenance

Request A

Admitted to the prompt Request B

Stopped before inference Source file is missing a date. No prompt is built, no tokens are spent.

Identity, scope and evidence are checked by rules, not by the model. If a required check fails, the request stops there.

Pass it. Block it. Mask it.

Pass Order history, last 90 days Sent within permitted scope

Block Internal pricing notes Excluded from the request

Mask Customer ID 4471-2209[CUSTOMER_17] Model sees a placeholder only

CUSTOMER_17 = 4471-2209The mapping is secured in your environment and restored only for authorized readers.

Explicit rules decide what passes, what is blocked and what is masked. Sensitive values never leave your control.

Only then do we build the prompt.

Governed prompt assembled by rules

The permitted task and answer requirements
Admitted evidence and relevant business rules
Validated calculations, with uncertainty stated
Required citations, format and limits

All required checks passed before submission

LLM

Writes a clear explanation
in an organized format

Governed tools do the math. The model does the writing. It does not get to rewrite the rules.

Then we check the work.

Run record versioned, replayable

Evidence	Sources, dates and evidence states
Calculations	Method, assumptions and uncertainty
Governance	Rules applied and permitted scope
Validation	Output checks and open gaps
Traceability	Prompt, model and required approvals

A useful answer, with its limits in plain view.

Evidence gaps stay visible
Output permissions applied
Approvals before consequential action

Every answer ships with its receipts. You can inspect the sources, the math and the rules behind it.

Govern the work. Then generate the answer.

Your data. Your rules. Your choice of model. Bring one real question from your business. We will walk it through a governed run with you, checks, receipts and all.

Book a call Read how it works

The standard against which your AI is measured.

ChosenClarityConfirmed